Overview of The New AI Org Chart
Nathaniel Whittemore explores how AI agents are changing not just how individuals work, but how organizations are structured. The episode centers on Jack Dorsey and Sequoia’s essay on Block’s attempt to replace traditional hierarchy with an AI-driven “company as intelligence,” then contrasts that top-down theory with a real-world, bottom-up example from Every, where employees and their agents are already forming a parallel org chart. The core argument: AI is beginning to take over the information-routing and coordination function that middle management has historically provided.
The Main Thesis: AI Is Changing the Org Chart, Not Just Productivity
The episode argues that most companies think of AI as a tool for making existing work faster, but the deeper shift is organizational:
- Traditional hierarchies exist to route information across large organizations.
- Middle management’s main job has been to aggregate context, align teams, and pass decisions up and down.
- AI agents may now be able to do some of that coordination work directly, reducing the need for human layers.
The show frames this as a historical inflection point: just as the Roman army, Prussian military reforms, railroads, and scientific management shaped modern org charts, AI could force a new structure centered on machine intelligence rather than human hierarchy.
Block’s Vision: The Company as an Intelligence System
The first half of the episode summarizes Block’s essay co-written by Jack Dorsey and Sequoia’s Roelof Botha. Their argument is that Block is trying to become something like a “mini-AGI” company, where AI replaces the coordination role of management.
Historical Framing
The essay traces organizational design through history:
- Roman military hierarchy established span-of-control logic.
- Prussian general staff formalized staff vs. line functions.
- Railroads brought military-style organization into business.
- Frederick Taylor’s scientific management optimized work inside the hierarchy.
- The matrix organization and later frameworks like McKinsey’s 7S tried to solve complexity and rigidity.
The key takeaway is that organizations have long been constrained by how much information humans can route and manage.
Block’s Proposed Model
Block’s vision is to replace hierarchy with two central models:
- A company world model: a continuously updated understanding of what’s happening internally.
- A customer world model: a data-rich understanding of customers and merchants, built from transaction data.
From those, AI can drive:
-
Capabilities
Core building blocks like payments, lending, card issuance, banking, payroll, etc. -
World models
A live understanding of the company and customers. -
An intelligence layer
AI that composes capabilities into proactive solutions for specific customer situations. -
Interfaces
Products like Square, Cash App, Afterpay, Tidal, and Bitkey that deliver the intelligence to users.
The Organizational Implication
Block’s model suggests a new staffing structure:
- Individual contributors build and maintain capabilities and models.
- DRIs (directly responsible individuals) own cross-functional outcomes.
- Player-coaches combine hands-on work with mentoring.
The point is to eliminate permanent middle management as an information-routing layer, because the system itself becomes the coordinator.
Every’s Experience: A Bottom-Up Parallel Org Chart
The second half of the episode compares Block’s theory with a recent discussion from Every’s podcast AI and I. Every represents the opposite direction: instead of redesigning the organization from above, it is discovering a new structure from the ground up through daily AI use.
Key Observations from Every
1. Agents naturally form a shadow org chart
As employees use personal agents over time, those agents begin to mirror the person’s specialization:
- A growth leader’s agent becomes the “go-to” for growth questions.
- A product leader’s agent becomes the best place for bug reports or feature requests.
This wasn’t designed intentionally; it emerged from repeated interactions. The episode calls this “compound engineering”—the agent accumulates a person’s working style and expertise over time.
2. Ownership creates trust
Every highlights a subtle but important distinction:
- Generic AI belongs to everyone
- A personal agent belongs to you
That ownership matters because the human is reputationally on the hook for what their agent does. If the agent answers badly in a shared Slack channel, the employee feels responsible. That skin in the game becomes a trust layer.
3. Public AI work teaches the organization
When agents work in public channels, everyone sees what’s possible. This creates a “mid-journey effect”:
- People learn what agents can actually do.
- The org becomes more willing to delegate.
- Capabilities spread through observation, not just documentation.
4. Group chat is still a weak point
One major limitation: current models are not great at group dynamics.
- Agents can get stuck in loops.
- They may “talk too much” and trigger each other.
- The transcript describes this as an “ant death spiral” in Slack.
This is a practical reminder that the social layer of organizational work is still hard for AI.
5. The biggest barrier is often imagination, not technology
Many useful workflows already worked technically, but people didn’t think to use them that way.
- Example: asking an agent to call and walk through emails during a commute.
- The ability was there; the user simply hadn’t mentally modeled the use case.
That suggests adoption depends heavily on building new habits of delegation.
6. Sharing capabilities across the org is unresolved
A key open question is how one person’s agent skills become useful to everyone else.
- Do you share skill files?
- Do skills stay personal?
- How do you onboard people into a company with dozens or hundreds of specialized agents?
This is less a technical issue than an organizational design problem.
Where Block and Every Agree
Despite their differences, both examples converge on a few important ideas:
- Middle management’s information-routing role is vulnerable first.
- AI can reduce the need for repeated human handoffs.
- Organizations will become faster when intelligence is closer to the edge.
- The org chart will likely change, even if the final form is still unclear.
Both imply that AI does not simply make existing structures more efficient—it changes what structures are needed at all.
Where They Differ
The episode draws a useful contrast:
Block: centralized intelligence
- One unified world model
- AI as the replacement for hierarchy
- A more architected, top-down rethinking of the whole company
Every: distributed intelligence
- Each employee’s agent reflects their own judgment and style
- Trust comes from personal ownership and reputation
- The organization becomes more like a network of specialized human-agent pairs
In other words, Block imagines a single coordinating intelligence, while Every demonstrates a distributed intelligence layer built around individuals.
Key Takeaways
- AI is increasingly being viewed as a replacement for the coordination function of management, not just a productivity tool.
- The most likely first casualty is the classic middle-manager role as information router.
- There are two emerging models:
- Top-down AI-native organizational redesign (Block)
- Bottom-up agent adoption with shadow org charts (Every)
- The hardest problems are now organizational, not purely technical:
- trust
- ownership
- group coordination
- onboarding
- sharing agent capabilities
- The future org chart may be a hybrid of centralized models and distributed human-agent ownership.
Bottom Line
The episode makes the case that AI will reshape organizations by changing how information flows. Block offers a bold theoretical blueprint for a company organized around machine intelligence, while Every shows the messy real-world process of agents becoming part of daily work. Together, they suggest that the future of work is not just “humans with better tools,” but a fundamental redesign of how companies coordinate, decide, and execute.